Archive · 23 Sep 2026
AI’s Uneven Intelligence Creates Enterprise Risks and Opportunities
AI can be excellent at some jobs and surprisingly poor at others.
The article says advanced systems can win math competitions but read an analog clock correctly only about half the time.
Companies are using AI more often, especially for writing, customer support and software development.
These tools can help people work faster and produce better results.
However, people can also make worse decisions when they use AI for tasks it does not understand well.
AI models are becoming cheaper, but companies may spend more because they use them more frequently.
Good company data helps AI give answers based on real records instead of guesses.
Businesses should test each AI task, watch its costs and errors, and involve a person when needed.
AI adoption rose sharply in 2025, but autonomous agents remained uncommon across most business functions.
Studies found productivity gains in customer support, writing, software development and suitable consulting tasks.
AI users performed worse on tasks outside a system’s capabilities, including one developer trial where work took longer.
Falling model prices are offset by rising usage, larger enterprise budgets and token-intensive agentic workflows.
The article recommends task-specific testing, data controls, cost monitoring and clear handoffs to human workers.
- Who
- Businesses, their employees and the AI systems they deploy are the main actors discussed.
- What
- The article examines AI’s uneven performance, productivity benefits, risks, costs and requirements for reliable enterprise deployment.
- When
- The article cites adoption and incident data from 2025, budget data through 2026, and a forecast through 2027.
- Where
- The discussion concerns enterprise workplaces and business functions, with examples from customer support, consulting, software development and other workflows.
- Why
- Companies are deploying AI to improve productivity, but must manage task-specific failures, data quality, rising usage costs and unclear accountability.
AI’s Business Potential
AI’s Operational Limits
Productivity
AI’s Business Potential
Research cited in the article found meaningful gains in customer support, writing, software development and consulting tasks within AI capabilities.
AI’s Operational Limits
The results are mixed and directional; gains disappear or reverse when tasks fall outside the system’s capabilities.
Developer assistance
AI’s Business Potential
AI-assisted development can improve workflow outcomes; one organisation reported roughly 30% fewer defects where data and processes were ready.
AI’s Operational Limits
A 2025 METR trial found experienced open-source developers took 19% longer with AI tools while believing they had worked faster.
Deployment and spending
AI’s Business Potential
Falling model prices and retrieval from company data create opportunities to scale useful AI services.
AI’s Operational Limits
Higher usage, token-intensive agents, rising budgets, weak controls and uncertain business value can make deployments costly or lead to cancellation.
- AI adoption
- 88% of organisations used AI in at least one business function in 2025, up from 78% in 2024.
- Generative AI use
- Generative AI was used by 70% of organisations in 2025.
- Customer-support productivity
- One study of 5,179 agents found an average productivity increase of 14%, with a 34% gain for the least experienced agents.
- Consulting experiment
- In a Harvard–BCG experiment involving 758 consultants, AI increased task completion by 12.2% and speed by 25.1% on suitable tasks.
- AI incidents
- Documented AI incidents increased from 233 to 362 in 2025.
- Enterprise AI budgets
- The average enterprise AI budget rose from $1.2 million in 2024 to $7.0 million in 2026, according to the FinOps Foundation survey.
- Agentic project outlook
- Gartner forecast that more than 40% of agentic AI projects could be cancelled by the end of 2027.